Rong-Hua Li

—

Unclaimed · AI-structured profileOpenAlex

LLM multi-agent systems researcher building graph-structured topologies, memory co-evolution, and diagnostic benchmarks

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-10-15) and cover the author's entire public record, not just the papers listed here. Reproducibility signal comes from Evolve's own reviews.

Research profile

Output—

No reliable output metrics are available: the snapshot records h-index 0 and i10-index 0 with no aggregated publication count, most plausibly an author-disambiguation gap. This axis would measure cumulative publication volume across venues and years.

Impact—

Citation impact data is insufficient — h-index and i10-index are both recorded as 0, which almost certainly reflects incomplete OpenAlex author linkage rather than measured impact. This axis would capture citations, uptake, and influence of the work.

Focus8/100

The radar records a focus score of 8/100. The raw distribution behind it spans five topics across eight assignments — Multi-Agent Systems and Negotiation (3), AI-based Problem Solving and Planning (2), and one each for topic modeling, NLP techniques, and reinforcement learning in robotics — while the field split is two-thirds LLM-based multi-agent systems and one-third their evaluation.

Collaboration54/100

Collaboration scores 54/100, based on 14 distinct coauthors in the snapshot. The same circle recurs across all three 2026 papers — Minghao An, Xunkai Li, Zekai Chen and Kairui Yang appear repeatedly, with Ziheng Yi, Kaixiang Zhang, Yuxuan Ba and Zhanke Liu also present. Rong-Hua Li sits in the last author position on every cataloged paper, a pattern consistent with a senior or supervisory role.

Activity40/100

Activity scores 40/100. The snapshot's counts-by-year contains a single entry — 3 works in 2026 — and recent3yWorks is likewise 3, meaning all cataloged output falls in one year with nothing earlier captured. This concentrated burst suggests a newly active or newly linked profile; the moderate score reflects the short observable window rather than a slowdown.

Reproducibility55/100

Reproducibility scores 55/100, the mean across all three papers with reproducibility data. Evolve rates each as medium readiness, and its verdicts consistently flag methodological caveats: single-run evaluations on MAGIC and MACE, sub-0.1 percentage-point margins over the SAGE baseline, and intervention evidence for OpenMAS-GCom covering only five of 29 datasets.

Research summary

Rong-Hua Li is a researcher working on LLM-based multi-agent systems, with an Evolve catalog centered on graph-structured methods for organizing and evaluating multi-agent collaboration. Three 2026 arXiv papers are on record, in all of which Li holds the last author position — a placement typically associated with senior supervision. Two are system papers: MAGIC introduces mixed-granularity agent graphs built incrementally with dense-reward reinforcement learning, and MACE couples agent adaptation with adaptive memory graphs for experience reuse. The third, OpenMAS-GCom, is a diagnostic benchmark applying controlled interventions to graph-enhanced multi-agent systems across 29 datasets. The split mirrors this profile's field distribution: two-thirds machine learning on LLM-based multi-agent systems, one-third on their evaluation. Topic data places Multi-Agent Systems and Negotiation first (3 mentions), with secondary presence in AI-based problem solving and planning, topic modeling, NLP techniques, and RL in robotics. Collaboration is moderate (54/100) across 14 distinct coauthors, with Minghao An, Xunkai Li, Zekai Chen and Kairui Yang recurring. Activity (40/100), output and impact reflect a snapshot limited to 2026, with h-index and i10-index recorded as 0 — most plausibly an author-linkage gap. Reproducibility averages 55/100, with Evolve flagging single-run evaluations and narrow margins on all three papers.

Collaboration pattern

Collaboration is concentrated in one stable circle: among 14 distinct coauthors, Minghao An, Xunkai Li, Zekai Chen and Kairui Yang recur across the 2026 papers, joined by Ziheng Yi, Kaixiang Zhang, Yuxuan Ba and Zhanke Liu. Rong-Hua Li is last author on all three cataloged papers and corresponding author on none, a signature of a supervisory rather than hands-on role. The collaboration score is 54/100.

Research keywords

LLM multi-agent systemsagent topology generationmixed granularityreinforcement learningreward shapingagent memorymemory graphsexperience reusegraph-enhanced MAScommunication topologycontrolled-intervention benchmarkingrole assignment

Highlights

MAGIC: RL-shaped construction of mixed-granularity agent graphs
Evolve's verdict: MAGIC learns per-role atomic-or-group multi-agent topologies via shaped reinforcement learning, ranking first on 8 benchmarks — a sound idea, though single-run results call for caution.
MACE: memory-agent co-evolution tops benchmarks on thin margins
Evolve's verdict: MACE's memory-graph co-evolution leads 8 benchmarks at 81.11% versus SAGE's 78.97%, but single-run evaluations, sub-0.1pp margins and uncited baselines leave its SOTA claim unproven.
OpenMAS-GCom: a controlled-intervention diagnostic for graph-enhanced MAS
Evolve's verdict: the benchmark claims controlled-intervention diagnosis of graph-enhanced MAS components across 29 datasets; the design is rigorous, but intervention evidence currently spans only five datasets.

Timeline

2026MAGIC appears on arXiv, introducing mixed-granularity agent graphs built incrementally with dense-reward reinforcement learning; Evolve reports first-place results on 8 benchmarks with a caution on single-run evaluation.
2026MACE presents memory-agent co-evolution with adaptive memory graphs, reporting 81.11% versus SAGE's 78.97% over 8 benchmarks; Evolve notes sub-0.1 percentage-point margins and single-run evaluations.
2026OpenMAS-GCom releases a diagnostic benchmark for graph-enhanced multi-agent systems spanning 29 datasets, with controlled-intervention evidence so far covering five of them.

Papers in the Evolve catalog

Frequent co-authors

Minghao An×3 Xunkai Li×3 Zekai Chen×3 Kairui Yang×2 Xunkai Li×2 Zekai Chen×2 Rong-Hua Li×2 Kairui Yang×2 Ziheng Yi×2 Kaixiang Zhang×1 Yuxuan Ba×1 Zhanke Liu×1

About this profile

This profile is AI-structured by Evolve.Pub from public bibliographic data (OpenAlex) and Evolve's own paper reviews. It is not written or endorsed by the researcher unless the profile is claimed. Radar dimensions are computed from real bibliometrics with fixed formulas — raw figures are shown alongside; missing dimensions mean insufficient data, never zero. Errors? Claim the profile to correct or hide it.

Bibliometric scores are computed from OpenAlex data only; h-index and i10-index are recorded as 0, most plausibly reflecting incomplete author disambiguation, so citation impact should be read as missing rather than zero.
Identity is name-resolved from public bibliographic data; institution, country and ORCID are absent, and confusion with other researchers named Rong-Hua Li cannot be ruled out.
All cataloged papers are 2026 arXiv preprints reviewed by Evolve; no peer-reviewed venue information is within scope.
The frequent-coauthor list contained duplicate entries and the researcher's own name; duplicates were removed before reporting.